CN108627241B - Dolphin widescriptae click signal detection method based on Gaussian mixture model - Google Patents

Dolphin widescriptae click signal detection method based on Gaussian mixture model Download PDF

Info

Publication number
CN108627241B
CN108627241B CN201810447762.4A CN201810447762A CN108627241B CN 108627241 B CN108627241 B CN 108627241B CN 201810447762 A CN201810447762 A CN 201810447762A CN 108627241 B CN108627241 B CN 108627241B
Authority
CN
China
Prior art keywords
signal
time
frequency
gaussian mixture
energy intensity
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201810447762.4A
Other languages
Chinese (zh)
Other versions
CN108627241A (en
Inventor
金海燕
张萌
肖照林
李秀秀
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Xi'an Jutong Elevator Co.,Ltd.
Original Assignee
Xian University of Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Xian University of Technology filed Critical Xian University of Technology
Priority to CN201810447762.4A priority Critical patent/CN108627241B/en
Publication of CN108627241A publication Critical patent/CN108627241A/en
Application granted granted Critical
Publication of CN108627241B publication Critical patent/CN108627241B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01H—MEASUREMENT OF MECHANICAL VIBRATIONS OR ULTRASONIC, SONIC OR INFRASONIC WAVES
    • G01H17/00—Measuring mechanical vibrations or ultrasonic, sonic or infrasonic waves, not provided for in the preceding groups
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/23—Clustering techniques
    • G06F18/232—Non-hierarchical techniques
    • G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/28—Quantising the image, e.g. histogram thresholding for discrimination between background and foreground patterns

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Probability & Statistics with Applications (AREA)
  • Multimedia (AREA)
  • Measurement Of Velocity Or Position Using Acoustic Or Ultrasonic Waves (AREA)
  • Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)

Abstract

The invention discloses a method for detecting a dolphin click signal based on a Gaussian mixture model, which comprises the steps of carrying out time-frequency analysis on an acquired marine signal by using windowed Fourier transform to obtain an initial signal time-frequency diagram of a marine sound signal, so that the subsequent steps are simple and quick to realize; the soft clustering method of the Gaussian mixture model is used for carrying out secondary clustering on the marine sound signals, so that the flexibility of detection is improved; the detection result has high accuracy and good integrity, and can be analogized to be used for detecting sound signals of various dolphins.

Description

Dolphin widescriptae click signal detection method based on Gaussian mixture model
Technical Field
The invention belongs to the technical field of sound signal processing, and particularly relates to a dolphin click signal detection method based on a Gaussian mixture model.
Background
The time-frequency analysis technology is one of research contents of digital signal processing, is commonly used in the fields of signal processing, fault diagnosis, feature extraction and the like, and can clearly describe the relation of signal frequency changing along with time. The basic idea of short-time fourier transform is to apply a sliding time window (truncation) to the signal and perform fourier transform on the signal in the window to obtain frequency domain data of the signal. Windowing is essentially the process of multiplying the original time domain signal by a so-called window function, so that the multiplied signal seems to better meet the periodicity requirements of the fourier transform.
Clustering is part of unsupervised learning in machine learning, and is often associated with subject research such as data mining, principal component analysis and image segmentation. In "unsupervised learning," the labeled information of the training samples is unknown, and the goal is to reveal the intrinsic properties of the data by learning the unlabeled training samples, providing a basis for further data analysis. Clustering is the process of dividing a collection of physical or abstract objects into classes composed of similar objects. The cluster generated by clustering is a collection of a set of data objects that are similar to objects in the same cluster and distinct from objects in other clusters. Conventional cluster analysis algorithms include partition-based methods, hierarchy-based methods, density-based methods, grid-based methods, and model-based methods. The most common of these are the partition-based K-means clustering method and the K-center point clustering method, and also the model-based gaussian mixture model clustering method. The K-means clustering method and the K-center clustering method result in each data point being partitioned into a cluster, and the gaussian mixture model clustering method results in the probability of these data points being partitioned into each cluster. Therefore, the K-means clustering method and the K-center point clustering method are collectively called hard clustering, and the gaussian mixture model clustering method is called soft clustering.
The gaussian mixture model is a model formed by accurately quantizing objects by using a gaussian probability density function and decomposing one object into a plurality of objects based on the gaussian probability density function. Briefly, a gaussian mixture model is a linear combination of several gaussian probability density functions. If the difference between the target area and the background area of an image is large and the target area and the background area have a certain difference in gray level, the gray level histogram of the image is in a double-peak-valley shape, wherein one peak corresponds to the target and the other peak corresponds to the background. In the case of complex images, the histogram of the image is usually multimodal, i.e. corresponds to a plurality of gaussian probability density functions, i.e. a gaussian distribution in a gaussian mixture model. Gaussian mixture model clustering is the partitioning of data into different clusters by partitioning these gaussian distributions.
The binarization technology is one of the most commonly used technologies in the field of digital image processing, and is commonly used in the research fields of image preprocessing, edge detection, image segmentation and the like. The image binarization is to set the gray value of a pixel point with the gray value of the image in the range of 0 to 255 as 0 or 255, and remove the intermediate transition pixels, that is, the whole image has a non-black or white visual effect. The most common method for image binarization is to set a threshold value T, and divide the pixel points of the image into two parts by T, namely a target object set larger than T and a background set smaller than T. According to different threshold value selection modes, image binarization methods are divided into a fixed threshold value method and an adaptive threshold value method. As the name suggests, the fixed threshold method is to set a fixed threshold not to change with the change of the image, and the adaptive threshold method is to calculate an optimal threshold adaptively according to the gray level distribution of the image pixel points. Although the selection method of the binary threshold value is hundreds, no general method can achieve the effect of the most ideal image for various images.
Disclosure of Invention
The invention aims to provide a method for detecting a dolphin click signal based on a Gaussian mixture model, which can extract a clear and complete detection dolphin click signal.
The technical scheme adopted by the invention is that the method for detecting the dolphin click signal based on the Gaussian mixture model is implemented according to the following steps:
step 1, collecting sound signals in the ocean, performing joint analysis on time domain information and frequency domain information of the sound signals in the ocean by using a time-frequency analysis technology to obtain time-frequency data of the signals, making an initial signal time-frequency graph, and obtaining energy intensity according to the initial signal time-frequency graph;
step 2, selecting the energy intensity as a clustered sample space, and performing secondary clustering on the sample space by using a Gaussian mixture model clustering method to obtain two clustered clusters;
and 3, solving a self-adaptive binarization segmentation threshold value according to the two cluster clusters, comparing the energy intensity in the step 1 with the segmentation threshold value, displaying a binarization signal time-frequency diagram according to a comparison result, and obtaining a complete and clear dolphin click signal detection result according to the binarization signal time-frequency diagram.
The specific process of the step 1 is as follows:
step 1.1, collecting sound signals in the sea through a sound sensor, and reading the sound signals in the sea by using an audio function of MAT L AB;
step 1.2, adding a sliding Hanning window to a sound signal read into the sea to perform signal truncation to obtain a truncation signal;
step 1.3, performing Fourier transform on the truncated signal in each window to obtain a function related to time and frequency;
step 1.4, an initial signal time-frequency graph is made according to the function of time and frequency, the time-frequency graph takes the time and the frequency as coordinate axes, and each point on the graph represents the energy intensity of the sound signals in the ocean under the frequency at the time.
Step 1.1 the sound signal in the ocean is present in MAT L AB in the form of a matrix with the number of rows representing the number of sample points at which the sound signal in the ocean is read in and the number of columns being the number of audio channels of the sound signal.
The specific process of the step 1.2 is as follows: adding a Hanning window to the sound signal read into the ocean, wherein a windowed signal time domain expression is a time domain expression of a window function of an original signal, and the Hanning window moves 1/3 with the size of the window each time, namely, the Hanning window intercepts the sound signal read into the ocean each time;
wherein, the time domain expression of the window function is:
Figure BDA0001657634050000041
step 1.3 fourier transform formula as follows:
Figure BDA0001657634050000042
where, f (w) represents the energy intensity at frequency w, and f (t) represents the cutoff signal.
The specific process of the step 2 is as follows:
step 2.1, using energy intensity as training set D ═ x of Gaussian mixture model clustering1，x2，…，xmAnd (4) the training set D satisfies the definition of a Gaussian mixture model:
Figure BDA0001657634050000043
wherein, αiIn order to obtain a mixing factor,i.e., the ratio of each gaussian distribution of the constituent gaussian mixture components, and satisfies
Figure BDA0001657634050000051
Constraint of uiAs mean vector, ∑iIs a covariance matrix, uiAnd ∑iIs a parameter of a gaussian probability density function that constitutes a gaussian mixture component;
p(x|ui,∑i) Satisfying the definition of gaussian distribution:
Figure BDA0001657634050000052
step 2.2, initializing model parameters of Gaussian mixture distribution { (α)i,ui,∑i)|1≤i＜2}；
Step 2.3, setting circulation termination conditions: c ═ 1 and cycle termination conditions: the number of cycles n;
step 2.4, random variable zj∈ {1,2} represents sample x that generates training set DjWherein j is 1,2,.. m, sample x is calculatedjPosterior probability p generated by ith Gaussian distributionM(zj＝i|xj) It is denoted as γji(i ═ 1,2), i.e.
γji＝pM(zj＝i|xj) (1≤i＜2) (5)；
Figure BDA0001657634050000053
Step 2.5, x for random variables of sample set DjA posteriori probability p ofM(xj) And solving the maximum likelihood estimation, namely maximizing the log likelihood, as follows:
Figure BDA0001657634050000054
step 2.6, calculating a new mean vector u according to LL (D)i', new covariance matrix ∑i' New mixtureSum coefficient αi', parameter { (α)i,ui,∑i) I is more than or equal to 1 and less than or equal to 2, is updated to be { (α)i',ui',∑i')|1≤i≤2}；
Step 2.7, judging whether the circulation termination condition is met;
if yes, making cluster Ci＝Φ(1≤i≤2)；
If not, the cycle number C is equal to C +1, and the step 2.4 is skipped to continue the execution;
step 2.8, for random variable xjCalculating the cluster mark lambda according to the following formulajX is to bejDividing into corresponding cluster: cλj＝Cλj∪{xj}；
Figure BDA0001657634050000061
The specific process of the step 2.6 is as follows:
order to
Figure BDA0001657634050000062
Can obtain uiAnd ∑iThe calculation formula of (a) is as follows:
Figure BDA0001657634050000063
Figure BDA0001657634050000064
for mixing coefficient αiIn addition to satisfy
Figure BDA0001657634050000065
It is also required to satisfy αi≥0，
Figure BDA0001657634050000066
Thus introducing a Lagrange multiplier, then
Figure BDA0001657634050000067
The formula pair αiIs 0, then there is:
Figure BDA0001657634050000068
α are multiplied by each otheriWhen λ ═ m is found by summing all the components mixed, then:
Figure BDA0001657634050000071
the specific process of the step 3 is as follows:
step 3.1, obtaining two clustering clusters of the signal according to a Gaussian mixture model clustering method, solving the maximum value max of the clustering cluster with low energy intensity, and solving the minimum value min of the clustering cluster with high energy intensity;
step 3.2, calculating the average value average of max and min, wherein the average value is used as a self-adaptive binary segmentation threshold;
step 3.3, judging whether the energy intensity obtained in the step 1 is greater than a segmentation threshold value:
if the energy intensity is greater than the segmentation threshold, the gray value of the sample point is 255;
if the energy intensity is not greater than the segmentation threshold, the gray value of the sample point is 0;
and 3.4, displaying the binarized signal time-frequency diagram, and obtaining a complete and clear dolphin click signal detection result according to the binarized signal time-frequency diagram.
In the step 3.1, the cluster with low energy intensity is generated by ocean noise, and the cluster with high energy intensity is click signal of the dolphin.
The invention discloses a method for detecting a dolphin click signal based on a Gaussian mixture model, which has the beneficial effects that:
1. according to the method, the acquired marine signals are subjected to time-frequency analysis by using windowed Fourier transform to obtain an initial signal time-frequency diagram of the marine sound signals, so that the subsequent steps are simple and quick to realize;
2. the invention uses the soft clustering method of the Gaussian mixture model to perform secondary clustering on the marine sound signals, thereby increasing the flexibility of detection;
3. the invention has high accuracy and good integrity of the detection result, and can be analogized to be used for detecting sound signals of various dolphins.
Drawings
FIG. 1 is a flow chart of a method for detecting a dolphin click signal based on a Gaussian mixture model according to the present invention;
FIG. 2 is a flow chart of a time-frequency analysis method according to the present invention;
FIG. 3 is a flow chart of a Gaussian mixture model clustering method of the present invention;
FIG. 4 is a flow chart of the adaptive binarization method of the present invention;
FIG. 5 is a waveform of an acoustic signal in an ocean in accordance with the present invention;
FIG. 6 is a time-frequency diagram of the initial signal obtained in the present invention;
FIG. 7 is an exemplary diagram of a Gaussian mixture model in an embodiment of the invention;
FIG. 8 is a diagram illustrating an example of the detection result in the embodiment of the present invention.
Detailed Description
The present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.
The invention relates to a method for detecting a dolphin click signal based on a Gaussian mixture model, which is specifically implemented according to the following steps as shown in FIG. 1:
step 1, collecting sound signals in the ocean, performing joint analysis on time domain information and frequency domain information of the sound signals in the ocean by using a time-frequency analysis technology to obtain time-frequency data of the signals, making an initial signal time-frequency graph, and obtaining energy intensity according to the initial signal time-frequency graph;
the specific process is as follows:
step 1.1, as shown in fig. 2, acquiring a sound signal in the sea by a sound sensor, and reading the sound signal in the sea by using an audio function of MAT L AB;
the sound signals in the ocean exist in MAT L AB in a matrix form, the row number of the matrix represents the number of sampling sample points when the sound signals in the ocean are read in, and the column number is the number of audio channels of the sound signals;
step 1.2, adding a sliding Hanning window to a sound signal read into the sea to perform signal truncation to obtain a truncation signal;
adding a Hanning window to the sound signal read into the ocean, wherein a windowed signal time domain expression is a time domain expression of a window function of an original signal, and the Hanning window moves 1/3 with the size of the window each time, namely, the Hanning window intercepts the sound signal read into the ocean each time;
wherein, the time domain expression of the window function is:
Figure BDA0001657634050000091
step 1.3, performing Fourier transform on the truncated signal in each window to obtain a function related to time and frequency;
the fourier transform equation is as follows:
Figure BDA0001657634050000092
wherein, F (w) represents the energy intensity when the frequency is w, and f (t) is a truncation signal;
step 1.4, an initial signal time-frequency graph is made according to the function of time and frequency, the time-frequency graph takes the time and the frequency as coordinate axes, and each point on the graph represents the energy intensity of the sound signals in the ocean under the frequency at the time.
Step 2, selecting the energy intensity as a clustered sample space, and performing secondary clustering on the sample space by using a Gaussian mixture model clustering method to obtain two clustered clusters;
the specific process is as follows:
step 2.1, as shown in fig. 3, using the energy intensity as the training set D ═ x of the gaussian mixture model cluster1，x2，…，xmAnd (4) the training set D satisfies the definition of a Gaussian mixture model:
Figure BDA0001657634050000101
wherein, αiIs a mixing coefficient, i.e. the proportion of each Gaussian distribution constituting a Gaussian mixture of components, and satisfies
Figure BDA0001657634050000102
Constraint of uiAs mean vector, ∑iIs a covariance matrix, uiAnd ∑iIs a parameter of a gaussian probability density function that constitutes a gaussian mixture component;
p(x|ui,∑i) Satisfying the definition of gaussian distribution:
Figure BDA0001657634050000103
step 2.2, initializing model parameters of Gaussian mixture distribution { (α)i,ui,∑i)|1≤i＜2}；
Step 2.3, setting circulation termination conditions: c ═ 1 and cycle termination conditions: the number of cycles n;
step 2.4, random variable zj∈ {1,2} represents sample x that generates training set DjWherein j is 1,2,.. m, sample x is calculatedjPosterior probability p generated by ith Gaussian distributionM(zj＝i|xj) It is denoted as γji(i ═ 1,2), i.e.
γji＝pM(zj＝i|xj) (1≤i＜2) (5)；
Figure BDA0001657634050000104
Step 2.5, x for random variables of sample set DjA posteriori probability p ofM(xj) And solving the maximum likelihood estimation, namely maximizing the log likelihood, as follows:
Figure BDA0001657634050000111
step 2.6, calculating a new mean vector u according to LL (D)i', new covariance matrix ∑i' New mixing coefficient αi', parameter { (α)i,ui,∑i) I is more than or equal to 1 and less than or equal to 2, is updated to be { (α)i',ui',∑i')|1≤i≤2}；
The specific process is as follows:
order to
Figure BDA0001657634050000112
Can obtain uiAnd ∑iThe calculation formula of (a) is as follows:
Figure BDA0001657634050000113
Figure BDA0001657634050000114
for mixing coefficient αiIn addition to satisfy
Figure BDA0001657634050000115
It is also required to satisfy αi≥0，
Figure BDA0001657634050000116
Thus introducing a Lagrange multiplier, then
Figure BDA0001657634050000117
The formula pair αiIs 0, then there is:
Figure BDA0001657634050000118
α are multiplied by each otheriWhen λ ═ m is found by summing all the components mixed, then:
Figure BDA0001657634050000119
step 2.7, judging whether the circulation termination condition is met;
if yes, making cluster Ci＝Φ(1≤i≤2)；
If not, the cycle number C is equal to C +1, and step 2.4 is executed again;
step 2.8, for random variable xjCalculating the cluster mark lambda according to the following formulajX is to bejDividing into corresponding cluster: cλj＝Cλj∪{xj}；
Figure BDA0001657634050000121
Step 3, solving a self-adaptive binarization segmentation threshold value according to the two cluster clusters, comparing the energy intensity in the step 1 with the segmentation threshold value, displaying a binarization signal time-frequency diagram according to a comparison result, and obtaining a complete and clear dolphin click signal detection result according to the binarization signal time-frequency diagram;
the specific process is as follows:
step 3.1, as shown in fig. 4, obtaining two clustering clusters of the signal according to a Gaussian mixture model clustering method, calculating the maximum value max of the clustering cluster with low energy intensity, and calculating the minimum value min of the clustering cluster with high energy intensity;
the cluster with low energy intensity is generated by ocean noise, and the cluster with high energy intensity is click signal of the dolphin widescreen;
step 3.2, calculating the average value average of max and min, wherein the average value is used as a self-adaptive binary segmentation threshold;
step 3.3, judging whether the energy intensity obtained in the step 1 is greater than a segmentation threshold value:
if the energy intensity is greater than the segmentation threshold, the gray value of the sample point is 255;
if the energy intensity is not greater than the segmentation threshold, the gray value of the sample point is 0;
and 3.4, displaying the binarized signal time-frequency diagram, and obtaining a complete and clear dolphin click signal detection result according to the binarized signal time-frequency diagram.
Examples
The implementation process of the method for detecting the dolphin click signal based on the Gaussian mixture model clustering is described by taking the sound signal collected by the sound sensor as an example.
Step 1.1, collecting sound signals in the sea through a sound sensor, and reading the sound signals in the sea by using an audio function of MAT L AB;
the sound signals in the ocean exist in MAT L AB in a matrix form, the row number of the matrix represents the number of sampling sample points when the sound signals in the ocean are read in, and the column number is the number of audio channels of the sound signals;
step 1.2, adding a sliding Hanning window to a sound signal read into the sea to perform signal truncation to obtain a truncation signal;
adding a Hanning window to the sound signal read into the ocean, wherein a windowed signal time domain expression is a time domain expression of a window function of an original signal, and the Hanning window moves 1/3 with the size of the window each time, namely, the Hanning window intercepts the sound signal read into the ocean each time;
wherein, the time domain expression of the window function is:
Figure BDA0001657634050000131
step 1.3, performing Fourier transform on the truncated signal in each window to obtain a function related to time and frequency;
the fourier transform equation is as follows:
Figure BDA0001657634050000132
wherein, F (w) represents the energy intensity when the frequency is w, and f (t) is a truncation signal;
step 1.4, an initial signal time-frequency graph is made according to the function of time and frequency, the time-frequency graph takes the time and the frequency as coordinate axes, and each point on the graph represents the energy intensity of the sound signals in the ocean under the frequency at the time.
The waveform diagram of the read marine sound signal is shown in fig. 5, and the time-frequency diagram of the initial signal is shown in fig. 6.
Step 2.1, using energy intensity as training set D ═ x of Gaussian mixture model clustering1，x2，…，xmAnd (4) the training set D satisfies the definition of a Gaussian mixture model:
Figure BDA0001657634050000141
wherein, αiIs a mixing coefficient, i.e. the proportion of each Gaussian distribution constituting a Gaussian mixture of components, and satisfies
Figure BDA0001657634050000142
Constraint of uiAs mean vector, ∑iIs a covariance matrix, uiAnd ∑iIs a parameter of a gaussian probability density function that constitutes a gaussian mixture component;
p(x|ui,∑i) Satisfying the definition of gaussian distribution:
Figure BDA0001657634050000143
step 2.2, initializing model parameters of Gaussian mixture distribution { (α)i,ui,∑i)|1≤i＜2}；
Step 2.3, setting circulation termination conditions: c ═ 1 and cycle termination conditions: the number of cycles n;
step 2.4, random variable zj∈ {1,2} represents sample x that generates training set DjWherein j is 1,2,.. m, sample x is calculatedjPosterior probability p generated by ith Gaussian distributionM(zj＝i|xj) It is denoted as γji(i ═ 1,2), i.e.
γji＝pM(zj＝i|xj) (1≤i＜2) (5)；
Figure BDA0001657634050000151
Step 2.5, x for random variables of sample set DjA posteriori probability p ofM(xj) And solving the maximum likelihood estimation, namely maximizing the log likelihood, as follows:
Figure BDA0001657634050000152
step 2.6, calculating a new mean vector u according to LL (D)i', new covariance matrix ∑i' New mixing coefficient αi', parameter { (α)i,ui,∑i) I is more than or equal to 1 and less than or equal to 2, is updated to be { (α)i',ui',∑i')|1≤i≤2}；
The specific process is as follows:
order to
Figure BDA0001657634050000153
Can obtain uiAnd ∑iThe calculation formula of (a) is as follows:
Figure BDA0001657634050000154
Figure BDA0001657634050000155
for mixing coefficient αiIn addition to satisfy
Figure BDA0001657634050000156
It is also required to satisfy αi≥0，
Figure BDA0001657634050000157
Thus introducing a Lagrange multiplier, then
Figure BDA0001657634050000158
The formula pair αiIs 0, then there is:
Figure BDA0001657634050000161
α are multiplied by each otheriWhen λ ═ m is found by summing all the components mixed, then:
Figure BDA0001657634050000162
step 2.7, judging whether the circulation termination condition is met;
if yes, making cluster Ci＝Φ(1≤i≤2)；
If not, the cycle number C is equal to C +1, and step 2.4 is executed again;
step 2.8, for random variable xjCalculating the cluster mark lambda according to the following formulajX is to bejDividing into corresponding cluster: cλj＝Cλj∪{xj}；
Figure BDA0001657634050000163
The time-frequency data of the signal is divided into two clusters, a dolphin click signal cluster with higher energy intensity and an ocean noise and reverberation cluster with weaker energy intensity, and an exemplary graph of a gaussian mixture model with 3 cluster centers is shown in fig. 7 below.
Step 3.1, obtaining two clustering clusters of the signal according to a Gaussian mixture model clustering method, solving the maximum value max of the clustering cluster with low energy intensity, and solving the minimum value min of the clustering cluster with high energy intensity;
the cluster with low energy intensity is generated by ocean noise, and the cluster with high energy intensity is click signal of the dolphin widescreen;
step 3.2, calculating the average value average of max and min, wherein the average value is used as a self-adaptive binary segmentation threshold;
step 3.3, judging whether the energy intensity obtained in the step 1 is greater than a segmentation threshold value:
if the energy intensity is greater than the segmentation threshold, the gray value of the sample point is 255;
if the energy intensity is not greater than the segmentation threshold, the gray value of the sample point is 0;
and 3.4, displaying the binarized signal time-frequency diagram, and obtaining a complete and clear dolphin click signal detection result according to the binarized signal time-frequency diagram.
As can be seen from the results of detection in FIG. 8, the lower shorter lateral line is where the dolphin click signal is detected.
By the mode, the dolphin click signal detection method based on the Gaussian mixture model performs time-frequency analysis on the acquired marine signals by using windowed Fourier transform to obtain an initial signal time-frequency diagram of the marine sound signals, so that the subsequent steps are simple and rapid to implement; the soft clustering method of the Gaussian mixture model is used for carrying out secondary clustering on the marine sound signals, so that the flexibility of detection is improved; the detection result has high accuracy and good integrity, and can be analogized to be used for detecting sound signals of various dolphins.

Claims (7)

1. A method for detecting a dolphin click signal based on a Gaussian mixture model is characterized by comprising the following steps:
step 1, collecting sound signals in the ocean, performing joint analysis on time domain information and frequency domain information of the sound signals in the ocean by using a time-frequency analysis technology to obtain time-frequency data of the signals, making an initial signal time-frequency graph, and obtaining energy intensity according to the initial signal time-frequency graph;
step 2, selecting the energy intensity as a clustered sample space, and performing secondary clustering on the sample space by using a Gaussian mixture model clustering method to obtain two clustered clusters; the specific process of the step 2 is as follows:
step 2.1, using energy intensity as training set D ═ x of Gaussian mixture model clustering1，x2，…，xmAnd (4) the training set D satisfies the definition of a Gaussian mixture model:
Figure FDA0002443179470000011
wherein, αiIs a mixing coefficient, i.e. the proportion of each Gaussian distribution constituting a Gaussian mixture of components, and satisfies
Figure FDA0002443179470000012
Constraint of uiAs mean vector, ∑iIs a covariance matrix, uiAnd ∑iIs a parameter of a gaussian probability density function that constitutes a gaussian mixture component;
p(x|ui,∑i) Satisfying the definition of gaussian distribution:
Figure FDA0002443179470000013
step 2.2, initializing model parameters of Gaussian mixture distribution { (α)i,ui,∑i)|1≤i＜2}；
Step 2.3, setting circulation termination conditions: c ═ 1 and cycle termination conditions: the number of cycles n;
step 2.4, random variable zj∈ {1,2} represents sample x that generates training set DjWherein j is 1,2,.. m, sample x is calculatedjPosterior probability p generated by ith Gaussian distributionM(zj＝i|xj) It is denoted as γji(i ═ 1,2), i.e.
γji＝pM(zj＝i|xj) (1≤i＜2) (5)；
Figure FDA0002443179470000021
Step 2.5, x for random variables of sample set DjA posteriori probability p ofM(xj) And solving the maximum likelihood estimation, namely maximizing the log likelihood, as follows:
Figure FDA0002443179470000022
step (ii) of2.6 computing a new mean vector u from LL (D)i', new covariance matrix ∑i' New mixing coefficient αi', parameter { (α)i,ui,∑i) I is more than or equal to 1 and less than or equal to 2, is updated to be { (α)i',ui',∑i')|1≤i≤2}；
Step 2.7, judging whether the circulation termination condition is met;
if yes, making cluster Ci＝Φ(1≤i≤2)；
If not, the cycle number C is equal to C +1, and step 2.4 is executed again;
step 2.8, for random variable xjCalculating the cluster mark lambda according to the following formulajX is to bejDividing into corresponding cluster: cλj＝Cλj∪{xj}；
Figure FDA0002443179470000023
Step 3, solving a self-adaptive binarization segmentation threshold value according to the two cluster clusters, comparing the energy intensity in the step 1 with the segmentation threshold value, displaying a binarization signal time-frequency diagram according to a comparison result, and obtaining a complete and clear dolphin click signal detection result according to the binarization signal time-frequency diagram; the specific process of the step 3 is as follows:
step 3.1, obtaining two clustering clusters of the signal according to a Gaussian mixture model clustering method, solving the maximum value max of the clustering cluster with low energy intensity, and solving the minimum value min of the clustering cluster with high energy intensity;
step 3.2, calculating the average value average of max and min, wherein the average value is used as a self-adaptive binary segmentation threshold;
step 3.3, judging whether the energy intensity obtained in the step 1 is greater than a segmentation threshold value:
if the energy intensity is greater than the segmentation threshold, the gray value of the sample point is 255;
if the energy intensity is not greater than the segmentation threshold, the gray value of the sample point is 0;
and 3.4, displaying the binarized signal time-frequency diagram, and obtaining a complete and clear dolphin click signal detection result according to the binarized signal time-frequency diagram.
2. The method for detecting the dolphin click signal based on the Gaussian mixture model as claimed in claim 1, wherein the specific process of step 1 is as follows:
step 1.1, collecting sound signals in the sea through a sound sensor, and reading the sound signals in the sea by using an audio function of MAT L AB;
step 1.2, adding a sliding Hanning window to a sound signal read into the sea to perform signal truncation to obtain a truncation signal;
step 1.3, performing Fourier transform on the truncated signal in each window to obtain a function related to time and frequency;
step 1.4, an initial signal time-frequency graph is made according to the function of time and frequency, the time-frequency graph takes the time and the frequency as coordinate axes, and each point on the graph represents the energy intensity of the sound signals in the ocean under the frequency at the time.
3. The method as claimed in claim 2, wherein the acoustic signals in the sea of step 1.1 are in matrix form in MAT L AB, the number of rows in the matrix represents the number of sampling sample points when the acoustic signals in the sea are read in, and the number of columns is the number of audio channels of the acoustic signals.
4. The method for detecting the dolphin click signal based on the Gaussian mixture model as claimed in claim 2, wherein the step 1.2 comprises the following specific processes: adding a Hanning window to the sound signal read into the ocean, wherein a windowed signal time domain expression is a time domain expression of a window function of an original signal, and the Hanning window moves 1/3 with the size of the window each time, namely, the Hanning window intercepts the sound signal read into the ocean each time;
wherein, the time domain expression of the window function is:
Figure FDA0002443179470000041
5. the method for detecting the dolphin click signal based on the Gaussian mixture model as claimed in claim 2, wherein the Fourier transform formula in step 1.3 is as follows:
Figure FDA0002443179470000042
where, f (w) represents the energy intensity at frequency w, and f (t) represents the cutoff signal.
6. The method for detecting the dolphin click signal based on the Gaussian mixture model as claimed in claim 1, wherein the specific process of step 2.6 is as follows:
order to
Figure FDA0002443179470000043
Can obtain uiAnd ∑iThe calculation formula of (a) is as follows:
Figure FDA0002443179470000051
Figure FDA0002443179470000052
for mixing coefficient αiIn addition to satisfy
Figure FDA0002443179470000053
It is also required to satisfy αi≥0，
Figure FDA0002443179470000054
Thus introducing a Lagrange multiplier, then
Figure FDA0002443179470000055
The formula pair αiIs 0, then there is:
Figure FDA0002443179470000056
α are multiplied by each otheriWhen λ ═ m is found by summing all the components mixed, then:
Figure FDA0002443179470000057
7. the method as claimed in claim 1, wherein the cluster with low energy intensity is generated by ocean noise in step 3.1, and the cluster with high energy intensity is the click signal of the dolphin.
CN201810447762.4A 2018-05-11 2018-05-11 Dolphin widescriptae click signal detection method based on Gaussian mixture model Active CN108627241B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810447762.4A CN108627241B (en) 2018-05-11 2018-05-11 Dolphin widescriptae click signal detection method based on Gaussian mixture model

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810447762.4A CN108627241B (en) 2018-05-11 2018-05-11 Dolphin widescriptae click signal detection method based on Gaussian mixture model

Publications (2)

Publication Number Publication Date
CN108627241A CN108627241A (en) 2018-10-09
CN108627241B true CN108627241B (en) 2020-07-24

Family

ID=63692838

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810447762.4A Active CN108627241B (en) 2018-05-11 2018-05-11 Dolphin widescriptae click signal detection method based on Gaussian mixture model

Country Status (1)

Country Link
CN (1) CN108627241B (en)

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109541969B (en) * 2018-10-24 2021-04-23 深圳市亿兆互联技术有限公司 System and method for monitoring whales based on lora
CN110119974B (en) * 2019-05-17 2022-07-05 武汉众诚华鑫科技有限公司 delta-GMM clustering algorithm-based intelligent pushing method for mobile packages
CN111242145B (en) * 2020-04-28 2020-08-18 南京芯视界微电子科技有限公司 Histogram statistical method and system
CN113014875B (en) * 2021-02-24 2022-06-03 自然资源部第三海洋研究所 Dolphin sound image synchronous acquisition system
CN113109050A (en) * 2021-03-18 2021-07-13 重庆大学 Rolling bearing weak fault diagnosis method based on cyclic pulse

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6178261B1 (en) * 1997-08-05 2001-01-23 The Regents Of The University Of Michigan Method and system for extracting features in a pattern recognition system
CN101650430A (en) * 2008-12-26 2010-02-17 中国科学院声学研究所 System for acquiring dolphin's sound signal and method thereof
CN102324232A (en) * 2011-09-12 2012-01-18 辽宁工业大学 Method for recognizing sound-groove and system based on gauss hybrid models
JP2017112840A (en) * 2015-12-21 2017-06-29 キーオン株式会社 Animal repellent device
CN106952251A (en) * 2017-02-28 2017-07-14 华南理工大学 A kind of image significance detection method based on Adsorption Model

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6178261B1 (en) * 1997-08-05 2001-01-23 The Regents Of The University Of Michigan Method and system for extracting features in a pattern recognition system
CN101650430A (en) * 2008-12-26 2010-02-17 中国科学院声学研究所 System for acquiring dolphin's sound signal and method thereof
CN102324232A (en) * 2011-09-12 2012-01-18 辽宁工业大学 Method for recognizing sound-groove and system based on gauss hybrid models
JP2017112840A (en) * 2015-12-21 2017-06-29 キーオン株式会社 Animal repellent device
CN106952251A (en) * 2017-02-28 2017-07-14 华南理工大学 A kind of image significance detection method based on Adsorption Model

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
Using Gaussian mixture models to detect and classify dolphin whistle and pulses;Pablo Peso Parada et al;《The Journal of the Acoustical Society of America》;20140630;第135卷(第6期);第3372页左栏倒数第10行- 第3380页左栏倒数第11行，图1-10 *
宽吻海豚Click信号的时域滤波检测方法;杨武夷 等;《声学学报》;20170731;第42卷(第4期);第446页右栏倒数第8行-第449页右栏第4行，图1-7 *

Also Published As

Publication number Publication date
CN108627241A (en) 2018-10-09

Similar Documents

Publication Publication Date Title
CN108627241B (en) Dolphin widescriptae click signal detection method based on Gaussian mixture model
CN109345508B (en) Bone age evaluation method based on two-stage neural network
Guo et al. BARNet: Boundary aware refinement network for crack detection
CN102496023B (en) Region of interest extraction method of pixel level
Lategahn et al. Texture classification by modeling joint distributions of local patterns with Gaussian mixtures
CN101286229B (en) Sonar image self-adapting division method based on stratified MRF
CN103761726B (en) Block adaptive image partition method based on FCM
CN105809175A (en) Encephaledema segmentation method and system based on support vector machine algorithm
CN105913081A (en) Improved PCAnet-based SAR image classification method
CN113657491A (en) Neural network design method for signal modulation type recognition
CN101030297A (en) Method for cutting complexity measure image grain
CN110991547A (en) Image significance detection method based on multi-feature optimal fusion
CN109801305A (en) SAR image change detection based on depth capsule network
CN114067915A (en) scRNA-seq data dimension reduction method based on deep antithetical variational self-encoder
CN113033567A (en) Oracle bone rubbing image character extraction method fusing segmentation network and generation network
CN113269706A (en) Laser radar image quality evaluation method, device, equipment and storage medium
CN102663681A (en) Gray scale image segmentation method based on sequencing K-mean algorithm
CN102509308A (en) Motion segmentation method based on mixtures-of-dynamic-textures-based spatiotemporal saliency detection
CN104952071A (en) Maximum between-cluster variance image segmentation algorithm based on GLSC (gray-level spatial correlation)
CN116704208B (en) Local interpretable method based on characteristic relation
CN105160666A (en) SAR (synthetic aperture radar) image change detection method based on non-stationary analysis and conditional random field
CN109902690A (en) Image recognition technology
Akther et al. Detection of Vehicle's Number Plate at Nighttime using Iterative Threshold Segmentation (ITS) Algorithm
CN111127407A (en) Fourier transform-based style migration counterfeit image detection device and method
Scharfenberger et al. Image saliency detection via multi-scale statistical non-redundancy modeling

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant
TR01 Transfer of patent right

Effective date of registration: 20211230

Address after: 710000 10606, floor 6, unit 1, building 3, Huihao International Plaza, hi tech Zone, Xi'an, Shaanxi Province

Patentee after: Xi'an Jutong Elevator Co.,Ltd.

Address before: 710048 No. 5 Jinhua South Road, Shaanxi, Xi'an

Patentee before: XI'AN University OF TECHNOLOGY

TR01 Transfer of patent right